Business Context
A retail company has built a BART model to predict the potential lifetime value of new customers based on their initial demographic data and first-week purchase behavior. The goal is to score a new batch of customers to identify high-value prospects for a premium loyalty program.
Data Preparation
Create a training set with customer demographics and first-week spend, and a similar scoring set for new customers. A bartGauss model is trained to predict 'clv_actual'.
Copied!
| 1 | DATA mycas.customer_train; |
| 2 | call streaminit(123); |
| 3 | DO customer_id = 1 to 2000; |
| 4 | age = 20 + floor(rand('UNIFORM') * 45); |
| 5 | first_week_spend = 50 + rand('NORMAL', 150, 75); |
| 6 | num_visits = 1 + floor(rand('UNIFORM') * 10); |
| 7 | clv_actual = (first_week_spend * 2.5) + (age * 15) - (num_visits * 20) + rand('NORMAL', 0, 500); |
| 8 | OUTPUT; |
| 9 | END; |
| 10 | RUN; |
| 11 | |
| 12 | DATA mycas.customer_score; |
| 13 | call streaminit(456); |
| 14 | DO customer_id = 2001 to 2500; |
| 15 | age = 18 + floor(rand('UNIFORM') * 50); |
| 16 | first_week_spend = 60 + rand('NORMAL', 120, 60); |
| 17 | num_visits = 1 + floor(rand('UNIFORM') * 8); |
| 18 | OUTPUT; |
| 19 | END; |
| 20 | RUN; |
Étapes de réalisation
1
Train the BART model using bartGauss and save the model state.
Copied!
| 1 | PROC CAS; |
| 2 | bart.bartGauss / |
| 3 | TABLE='customer_train', |
| 4 | inputs={{name='age'}, {name='first_week_spend'}, {name='num_visits'}}, |
| 5 | target='clv_actual', |
| 6 | saveState={name='clv_model', replace=true}; |
| 7 | QUIT; |
2
Score the new customers using bartScore, generating custom-named 99% credible limits and copying customer identifiers to the output table.
Copied!
| 1 | PROC CAS; |
| 2 | bart.bartScore / |
| 3 | TABLE='customer_score', |
| 4 | restore='clv_model', |
| 5 | casOut={name='clv_predictions', replace=true}, |
| 6 | copyVars={'customer_id', 'age'}, |
| 7 | alpha=0.01, |
| 8 | pred='Predicted_CLV', |
| 9 | lcl='Lower_99_CLV', |
| 10 | ucl='Upper_99_CLV', |
| 11 | resid='Prediction_Error'; |
| 12 | QUIT; |
3
Verify the output table structure and content.
Copied!
| 1 | PROC CAS; |
| 2 | TABLE.columnInfo / TABLE='clv_predictions'; |
| 3 | TABLE.fetch / TABLE='clv_predictions', to=10; |
| 4 | QUIT; |
Expected Result
The output table 'mycas.clv_predictions' should be created. It must contain the 'customer_id' and 'age' columns from the input table. It must also contain the newly generated columns: 'Predicted_CLV', 'Lower_99_CLV', 'Upper_99_CLV', and 'Prediction_Error'. The values in these columns should be populated for all 500 customers from the scoring table.